Context.cpp

#include "vkcv/Context.hpp"
#include "vkcv/Window.hpp"
#include "vkcv/Core.hpp"

namespace vkcv
{
    Context::Context(Context &&other) noexcept:
            m_Instance(other.m_Instance),
            m_PhysicalDevice(other.m_PhysicalDevice),
            m_Device(other.m_Device),
			m_FeatureManager(std::move(other.m_FeatureManager)),
			m_QueueManager(std::move(other.m_QueueManager)),
			m_Allocator(other.m_Allocator)
    {
        other.m_Instance        = nullptr;
        other.m_PhysicalDevice  = nullptr;
        other.m_Device          = nullptr;
		other.m_Allocator		= nullptr;
    }

    Context & Context::operator=(Context &&other) noexcept
    {
        m_Instance          = other.m_Instance;
        m_PhysicalDevice    = other.m_PhysicalDevice;
        m_Device            = other.m_Device;
        m_FeatureManager	= std::move(other.m_FeatureManager);
        m_QueueManager		= std::move(other.m_QueueManager);
        m_Allocator			= other.m_Allocator;

        other.m_Instance        = nullptr;
        other.m_PhysicalDevice  = nullptr;
        other.m_Device          = nullptr;
        other.m_Allocator		= nullptr;

        return *this;
    }

    Context::Context(vk::Instance instance,
                     vk::PhysicalDevice physicalDevice,
                     vk::Device device,
                     FeatureManager&& featureManager,
					 QueueManager&& queueManager,
					 vma::Allocator&& allocator) noexcept :
    m_Instance(instance),
    m_PhysicalDevice(physicalDevice),
    m_Device(device),
    m_FeatureManager(std::move(featureManager)),
    m_QueueManager(std::move(queueManager)),
    m_Allocator(allocator)
    {}

    Context::~Context() noexcept
    {
    	m_Allocator.destroy();
        m_Device.destroy();
        m_Instance.destroy();
    }

    const vk::Instance &Context::getInstance() const
    {
        return m_Instance;
    }

    const vk::PhysicalDevice &Context::getPhysicalDevice() const
    {
        return m_PhysicalDevice;
    }

    const vk::Device &Context::getDevice() const
    {
        return m_Device;
    }
    
    const FeatureManager& Context::getFeatureManager() const {
    	return m_FeatureManager;
    }
    
    const QueueManager& Context::getQueueManager() const {
    	return m_QueueManager;
    }
    
    const vma::Allocator& Context::getAllocator() const {
    	return m_Allocator;
    }
	
	/**
	 * @brief The physical device is evaluated by three categories:
	 * discrete GPU vs. integrated GPU, amount of queues and its abilities, and VRAM.physicalDevice.
	 * @param physicalDevice The physical device
	 * @return Device score as integer
	*/
	int deviceScore(const vk::PhysicalDevice& physicalDevice)
	{
		int score = 0;
		vk::PhysicalDeviceProperties properties = physicalDevice.getProperties();
		std::vector<vk::QueueFamilyProperties> qFamilyProperties = physicalDevice.getQueueFamilyProperties();
		
		// for every queue family compute queue flag bits and the amount of queues
		for (const auto& qFamily : qFamilyProperties) {
			uint32_t qCount = qFamily.queueCount;
			uint32_t bitCount = (static_cast<uint32_t>(qFamily.queueFlags & vk::QueueFlagBits::eCompute) != 0)
								+ (static_cast<uint32_t>(qFamily.queueFlags & vk::QueueFlagBits::eGraphics) != 0)
								+ (static_cast<uint32_t>(qFamily.queueFlags & vk::QueueFlagBits::eTransfer) != 0)
								+ (static_cast<uint32_t>(qFamily.queueFlags & vk::QueueFlagBits::eSparseBinding) != 0);
			score += static_cast<int>(qCount * bitCount);
		}
		
		// compute the VRAM of the physical device
		vk::PhysicalDeviceMemoryProperties memoryProperties = physicalDevice.getMemoryProperties();
		auto vram = static_cast<int>(memoryProperties.memoryHeaps[0].size / static_cast<uint32_t>(1E9));
		score *= vram;
		
		if (properties.deviceType == vk::PhysicalDeviceType::eDiscreteGpu) {
			score *= 2;
		}
		else if (properties.deviceType != vk::PhysicalDeviceType::eIntegratedGpu) {
			score = -1;
		}
		
		return score;
	}
	
	/**
	 * @brief All existing physical devices will be evaluated by deviceScore.
	 * @param instance The instance
	 * @param physicalDevice The optimal physical device
	 * @return Returns if a suitable GPU is found as physical device
	 * @see Context.deviceScore
	*/
	bool pickPhysicalDevice(const vk::Instance& instance, vk::PhysicalDevice& physicalDevice)
	{
		const std::vector<vk::PhysicalDevice>& devices = instance.enumeratePhysicalDevices();
		
		if (devices.empty()) {
			vkcv_log(LogLevel::ERROR, "Failed to find GPUs with Vulkan support");
			return false;
		}
		
		int max_score = -1;
		for (const auto& device : devices) {
			int score = deviceScore(device);
			if (score > max_score) {
				max_score = score;
				physicalDevice = device;
			}
		}
		
		if (max_score == -1) {
			vkcv_log(LogLevel::ERROR, "Failed to find a suitable GPU");
			return false;
		} else {
			return true;
		}
	}
	
	/**
	 * @brief Check whether all string occurrences in "check" are contained in "supported"
	 * @param supported The const vector const char* reference used to compare entries in "check"
	 * @param check The const vector const char* reference elements to be checked by "supported"
	 * @return True, if all elements in "check" are supported (contained in supported)
	*/
	bool checkSupport(const std::vector<const char*>& supported, const std::vector<const char*>& check)
	{
		for (auto checkElem : check) {
			bool found = false;
			for (auto supportedElem : supported) {
				if (strcmp(supportedElem, checkElem) == 0) {
					found = true;
					break;
				}
			}
			if (!found)
				return false;
		}
		return true;
	}
	
	std::vector<std::string> getRequiredExtensions() {
		std::vector<std::string> extensions = Window::getExtensions();
		
#ifndef NDEBUG
		extensions.emplace_back(VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
#endif
		
		return extensions;
	}

	Context Context::create(const char *applicationName,
							uint32_t applicationVersion,
							const std::vector<vk::QueueFlagBits>& queueFlags,
							const Features& features,
							const std::vector<const char*>& instanceExtensions) {
		// check for layer support
		
		const std::vector<vk::LayerProperties>& layerProperties = vk::enumerateInstanceLayerProperties();
		
		std::vector<const char*> supportedLayers;
		supportedLayers.reserve(layerProperties.size());
		
		for (auto& elem : layerProperties) {
			supportedLayers.push_back(elem.layerName);
		}

// if in debug mode, check if validation layers are supported. Enable them if supported
#ifndef NDEBUG
		std::vector<const char*> validationLayers = {
				"VK_LAYER_KHRONOS_validation"
		};
		
		if (!checkSupport(supportedLayers, validationLayers)) {
			throw std::runtime_error("Validation layers requested but not available!");
		}
#endif
		
		// check for instance extension support
		std::vector<vk::ExtensionProperties> instanceExtensionProperties = vk::enumerateInstanceExtensionProperties();
		
		std::vector<const char*> supportedExtensions;
		supportedExtensions.reserve(instanceExtensionProperties.size());
		
		for (auto& elem : instanceExtensionProperties) {
			supportedExtensions.push_back(elem.extensionName);
		}
		
		// for GLFW: get all required extensions
		auto requiredStrings = getRequiredExtensions();
		std::vector<const char*> requiredExtensions;
		
		for (const auto& extension : requiredStrings) {
			requiredExtensions.push_back(extension.c_str());
		}
		
		requiredExtensions.insert(requiredExtensions.end(), instanceExtensions.begin(), instanceExtensions.end());
		
		if (!checkSupport(supportedExtensions, requiredExtensions)) {
			throw std::runtime_error("The requested instance extensions are not supported!");
		}
		
		const vk::ApplicationInfo applicationInfo(
				applicationName,
				applicationVersion,
				VKCV_FRAMEWORK_NAME,
				VKCV_FRAMEWORK_VERSION,
				VK_HEADER_VERSION_COMPLETE
		);
		
		vk::InstanceCreateInfo instanceCreateInfo(
				vk::InstanceCreateFlags(),
				&applicationInfo,
				0,
				nullptr,
				static_cast<uint32_t>(requiredExtensions.size()),
				requiredExtensions.data()
		);

#ifndef NDEBUG
		instanceCreateInfo.enabledLayerCount = static_cast<uint32_t>(validationLayers.size());
		instanceCreateInfo.ppEnabledLayerNames = validationLayers.data();
#endif
		
		vk::Instance instance = vk::createInstance(instanceCreateInfo);

		std::vector<vk::PhysicalDevice> physicalDevices = instance.enumeratePhysicalDevices();
		vk::PhysicalDevice physicalDevice;
		
		if (!pickPhysicalDevice(instance, physicalDevice)) {
			throw std::runtime_error("Picking suitable GPU as physical device failed!");
		}
		
		FeatureManager featureManager (physicalDevice);
		
#ifdef __APPLE__
		featureManager.useExtension("VK_KHR_portability_subset", true);
#endif

		if (featureManager.useExtension(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME, false)) {
			featureManager.useFeatures<vk::PhysicalDeviceShaderFloat16Int8Features>(
					[](vk::PhysicalDeviceShaderFloat16Int8Features& features) {
				features.setShaderFloat16(true);
			}, false);
		}
		
		if (featureManager.useExtension(VK_KHR_16BIT_STORAGE_EXTENSION_NAME, false)) {
			featureManager.useFeatures<vk::PhysicalDevice16BitStorageFeatures>(
					[](vk::PhysicalDevice16BitStorageFeatures& features) {
				features.setStorageBuffer16BitAccess(true);
			}, false);
		}

		featureManager.useFeatures([](vk::PhysicalDeviceFeatures& features) {
			features.setFragmentStoresAndAtomics(true);
			features.setGeometryShader(true);
			features.setDepthClamp(true);
			features.setShaderInt16(true);
		});
		
		for (const auto& feature : features.getList()) {
			feature(featureManager);
		}
		
		const auto& extensions = featureManager.getActiveExtensions();

		std::vector<vk::DeviceQueueCreateInfo> qCreateInfos;
		std::vector<float> qPriorities;
		qPriorities.resize(queueFlags.size(), 1.f);
		std::vector<std::pair<int, int>> queuePairsGraphics, queuePairsCompute, queuePairsTransfer;
		
		QueueManager::queueCreateInfosQueueHandles(physicalDevice, qPriorities, queueFlags, qCreateInfos, queuePairsGraphics, queuePairsCompute, queuePairsTransfer);
		
		vk::DeviceCreateInfo deviceCreateInfo(
				vk::DeviceCreateFlags(),
				qCreateInfos.size(),
				qCreateInfos.data(),
				0,
				nullptr,
				extensions.size(),
				extensions.data(),
				nullptr
		);

#ifndef NDEBUG
		deviceCreateInfo.enabledLayerCount = static_cast<uint32_t>(validationLayers.size());
		deviceCreateInfo.ppEnabledLayerNames = validationLayers.data();
#endif

		deviceCreateInfo.setPNext(&(featureManager.getFeatures()));
		
		vk::Device device = physicalDevice.createDevice(deviceCreateInfo);

		if (featureManager.isExtensionActive(VK_NV_MESH_SHADER_EXTENSION_NAME)) {
			InitMeshShaderDrawFunctions(device);
		}
		
		QueueManager queueManager = QueueManager::create(
				device,
				queuePairsGraphics,
				queuePairsCompute,
				queuePairsTransfer
		);

		vma::AllocatorCreateFlags vmaFlags;
		const vma::AllocatorCreateInfo allocatorCreateInfo (
				vmaFlags,
				physicalDevice,
				device,
				0,
				nullptr,
				nullptr,
				nullptr,
				nullptr,
				instance,
				VK_HEADER_VERSION_COMPLETE
		);
		
		vma::Allocator allocator = vma::createAllocator(allocatorCreateInfo);
		
		return Context(
				instance,
				physicalDevice,
				device,
				std::move(featureManager),
				std::move(queueManager),
				std::move(allocator)
		);
	}
 
}